One API call puts intelligent voice agents on any line — agents that speak, listen, and answer from your organisation's own knowledge, with every word audited. Built for banking, insurance, telco and government.
Deploys in your VPC · POST /v1/calls to first live call in under an hour
Hi — there's a charge on my card I don't recognise. It's on the card ending •••• 4412 — redacted
PAN detected → masked in transcript, audio scrubbed. PCI event logged.
I can help with that. I've flagged the transaction as disputed — you're covered while we investigate, and a temporary credit is applied within 2 business days.
⌕ grounded · Card Disputes Policy v3.2 §4.1
That's great. Will I get a new card?
Yes — I've ordered a replacement to your registered address. It arrives in 3–5 days; your digital wallet keeps working in the meantime.
VoiceKernel is not a bot builder. It's the layer your engineers call and your risk team signs off on.
Run multiple agents on a single call — a speaker, a compliance listener, a learner capturing intelligence. Attach them to AI-led calls or live human conversations.
Answers come from your knowledge — policies, product docs, CRM, core systems — with per-utterance citations, so every response is explainable after the fact.
SIP, PSTN and WebRTC. Native handoff to Genesys, NICE and Amazon Connect. Webhooks into your CRM so the voice channel writes back, not just talks.
PCI redaction in-stream, consent capture, retention policies, data residency, full reasoning logs. Deploy in our cloud, your VPC, or on-prem.
Live call monitoring, sentiment, containment rate, escalation triggers, and A/B testing across agent versions — from one console your CX team actually uses.
If you've outgrown developer-toy voice platforms, this is the layer you migrate to. Same API ergonomics, enterprise everything else.
Most platforms put one bot on one call. VoiceKernel treats a call as a room: agents join with distinct roles and permissions, whether the other side is a customer, an IVR, or your own human team.
Talks to the customer. Grounded in your knowledge base, constrained by your guardrails, tuned per queue — disputes, claims, renewals, hardship.
Silent on the line. Redacts card numbers in-stream, monitors for vulnerable-customer signals, and can force escalation to a human in real time.
Turns every call into structured data — intents, objections, product gaps — and feeds it back to your CRM and analytics without a single manual QA review.
Everything is an API. Create agents, attach knowledge, place or receive calls, stream transcripts — with a TypeScript SDK, and OpenAPI to generate a client for everything else.
Prototype in minutes on our cloud, then flip the same config to your VPC for production. No re-platforming between pilot and rollout.
Read the docs# place an outbound call with a grounded agent + compliance listener curl https://api.voicekernel.ai/v1/calls \ -H "Authorization: Bearer $VK_KEY" \ -d '{ "to": "+61400000000", "agents": [ { "id": "agent_disputes", "role": "speak", "knowledge": "kb_cards_policies" }, { "id": "agent_pci_guard", "role": "listen", "actions": ["redact_pan", "escalate_vulnerable"] }, { "id": "agent_intel", "role": "learn", "sink": "webhook:crm_events" } ], "region": "au-syd", "record": { "consent": "prompt_first" } }'
The reason voice AI stalls in enterprise isn't the demo — it's the sign-off. VoiceKernel is designed for the sign-off.
Card numbers and identifiers are masked in audio and transcript before they ever reach storage. PCI events are logged, not hoped for.
Every agent utterance carries its source citation and reasoning log. When a regulator asks why the AI said it, you have the answer.
Region-pinned processing and storage — Australia, EU, US — with VPC and on-prem deployment for teams that can't send audio offshore.
Confidence thresholds, forced escalation rules and live takeover. The AI handles volume; your people handle judgement.
A 30-minute technical demo on your own use case — bring a policy document and we'll show you a grounded agent answering from it, live.